以視覺辨識為基礎之停車場車位偵測;A Visual Recognition-Based Detection of Parking Slots

National Chung Cheng University Institutional Repository > 工學院 > 電機工程研究所 > 學位論文 > Item A095B0000Q/346

Please use this identifier to cite or link to this item: http://ccur.lib.ccu.edu.tw/handle/A095B0000Q/346

Title:	以視覺辨識為基礎之停車場車位偵測;A Visual Recognition-Based Detection of Parking Slots
Authors:	黃柏穎;Huang, Po-In
Contributors:	電機工程研究所
Keywords:	行車紀錄器;停車格;停車場;停車位;先進駕駛輔助系統;parking space;parking slot;parking;event data recorder;ADAS
Date:	2016
Issue Date:	2019-07-17
Publisher:	電機工程研究所
Abstract:	本論文提出以視覺為基礎的停車位偵測的設計及驗證，對於天候(日間、夜間)搭配多種不同停車場景之挑戰，我們的演算法採用了動態興趣區間可隨著行車狀況自動調整區間位置，方便適應不同場景之計算；而對於不同光線停車場景的挑戰，我們也採用不同的邊緣與雜訊過濾法，在畫面複雜之場景可去除更多多餘雜訊以利計算。本論文更整合了多種靜態偵測停車格偵測系統與改良之車道線偵測系統，建立實時動態偵測。於車體前方10至20公尺範圍內，車速限制20km/h之下，利用偵測所得之左、右車道主線，將其向外延伸，設置動態ROI位置後，配合視覺上的物理材質、車體陰影、T字圖像比對計算，以達到更高的準確度。本演算法實作於Visual Studio 2010，採用解析度1280×720之HD影像輸入，在此解析度下，視覺停車位偵測的處理效能可達到每秒30張。而實驗數據統計顯示，在停車偵測系統在日間的偵測率可以達到將近82%的準確度；但是停車偵測系統在夜間卻只能達到75%的偵測率，畢竟夜間的場景路燈的位置還是影響很多的。 This thesis proposed the design, and verification of Visual Recognition-Based Detection of Parking Slots (VBDS), which could be a new function of Advance Driver Assistance System (ADAS). Dynamic ROI is adopted to reduce the effect of various weather conditions. Adaptive Canny filter is also adopted to conquer the effect of some tough parking scenes. The proposed algorithm integrated variety of static parking detection methods and also combined an improved Lane Detection System(LDS) for parking scenes to establish a real-time detection. Within the range of 10 to 20 meters in front of the vehicles and driving speed 20km/h, by using the result of LDS, we found two main guideline, and setting dynamic ROI of both side. With the calculation of texture, shadow of vehicles, and T-shape template matching, we can achieve a good accuracy of experiment result. The proposed system is implemented on Visual Studio 2010, adopted an input video, which resolution is 1280×720 (HD).In this resolution, the performance of the proposed VBDS can achieve 30fps. Experiment shows that accuracy of the proposed VBDS is almost 82% at day-case. However the accuracy can only reach 75% at night-case due to the lights position of scenes.
Appears in Collections:	[電機工程研究所] 學位論文

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	417	View/Open

Loading...